The AI industry's transition from the training phase to the era of inference and agentic AI is shifting market focus back to the CPU, a component once overshadowed by the prominence of GPUs.
A recent report from UBS highlights that Agentic AI will propel server CPU demand to significantly exceed traditional cycles. In this evolving landscape, Advanced Micro Devices (AMD) is expanding its server market share with strong product competitiveness, while the ARM architecture is gaining favor with cloud and AI infrastructure providers due to its energy efficiency advantages. Concurrently, Bank of America's latest analysis suggests the expansion in CPU demand will concurrently drive growth in advanced manufacturing and packaging, benefiting leading manufacturers like Taiwan Semiconductor Manufacturing (TSM) and ASE Technology (ASX).
Shift in AI Infrastructure Focus
The shared foundation for both institutions' assessments is that AI infrastructure is evolving from a sole pursuit of computing power scale towards building more comprehensive computing systems. While GPUs remain responsible for core training and inference tasks, the role of CPUs in data management, agent execution, multi-model coordination, and system orchestration is becoming increasingly critical. Bank of America projects the global server CPU market to grow to over $170 billion by 2030.
As AMD and ARM continue to capture more market share, a greater volume of high-end CPU orders is expected to flow to advanced process suppliers like TSM. Meanwhile, the adoption of Chiplet architectures and rising complexity are also anticipated to benefit advanced packaging providers such as ASX.
Agentic AI Elevates CPU to Central Role
In recent years, AI infrastructure build-out has primarily centered on GPUs, with CPUs handling auxiliary roles like data scheduling, storage management, and task coordination. As AI applications shift from training to inference and further towards Agentic AI, the importance of CPUs is rapidly increasing.
In Agentic AI scenarios, models must plan tasks, invoke tools, access databases, manage states, and execute complex workflows. These tasks are not solely reliant on matrix computations but involve a significant number of sequential operations with low latency and high I/O requirements, making them more suitable for CPU processing.
CPUs are evolving from traditional "host processors" to become the "central control unit" of AI systems. If a CPU cannot efficiently handle data scheduling, memory management, and task coordination, even the most powerful GPU may suffer from utilization issues.
This trend is expected to drive the global server CPU market into a new growth cycle, with its size projected to surpass $170 billion by 2030, a nearly fivefold increase from approximately $35 billion in 2025.
Standalone AI Server Market Favors x86, AMD Seizes Opportunity
Amid the CPU demand surge, UBS is particularly optimistic about AMD's long-term growth potential.
The firm believes standalone Agentic AI servers are rapidly emerging. These applications prioritize core count and multi-threading performance, areas where AMD holds a distinct advantage. UBS's latest estimates suggest that in the standalone AI server market, x86 and ARM architectures will capture approximately 60% and 40% market share respectively, with AMD positioned to secure the majority of the incremental x86 demand.
Based on this higher market share expectation, UBS has raised its server CPU revenue forecasts for AMD to $23 billion for 2027 (up from $21 billion) and to $29 billion for 2028 (up from $27 billion). The firm also projects server CPU revenue could reach $50 billion by 2030, exceeding its previous forecast of $41 billion.
Consequently, UBS has significantly increased its price target for AMD to $670 from $455.
ARM Ecosystem Targets Dominant Share in Key Segment
In addition to AMD, ARM is another key focus for UBS.
The report notes that future AI infrastructure will likely feature two main CPU architecture categories: one for Head Node CPUs connected to GPUs, emphasizing low latency and high energy efficiency; and another for standalone Agentic AI servers, prioritizing core count and throughput.
UBS believes ARM architecture holds a clear advantage in the former market, with products like NVIDIA's Grace and Vera, Google's Axion, and AWS's Graviton rapidly expanding. The firm projects that by 2030, ARM architecture could capture about 70% of the Head Node CPU market and secure close to 50% of the overall server CPU revenue share.
Based on this optimistic outlook for AI CPUs, UBS has raised its price target for ARM Holdings (ARM) to $470 from $260.
New Expansion Cycle for Foundry and Packaging Leaders
The surge in CPU demand is emerging as a new growth engine for the semiconductor industry, following the GPU wave. Bank of America notes this trend benefits not only chip designers but also directly impacts manufacturing and packaging/testing segments.
According to Bank of America's calculations, the global semiconductor manufacturing market related to server CPUs is expected to expand from $15 billion in 2025 to $49 billion by 2028. Within this, the outsourced production share is projected to rise from 52% to 71%, reflecting the strengthening core position of pure-play foundries like TSM in the high-end CPU space.
The scarcity of advanced process capacity, combined with parallel volume growth from multiple customers and architectures, makes the foundry segment one of the most certain beneficiaries in this upturn.
Simultaneously, AI CPUs' rising demand for Chiplet heterogeneous integration, CoWoS advanced packaging, and system-level testing is driving rapid expansion in the backend packaging and testing market. Bank of America estimates the packaging and testing market related to server CPUs will grow from $1.9 billion in 2025 to $9.6 billion by 2028, increasing its share of the advanced packaging market from 11% to 24%.
This upgrade in the packaging value chain not only increases the packaging value per chip but also extends the processing time for backend procedures like testing, offering packaging leaders a dual benefit of volume and price increases.
Based on this supply-demand analysis, Bank of America has concurrently raised its valuation expectations for supply chain leaders like TSM and ASX, viewing advanced manufacturing and packaging as the segments with the highest barriers in the industry chain. In the new wave of computing architecture transformation driven by Agentic AI, the trend of parallel evolution for CPUs and GPUs is becoming clearer. Foundry and packaging/testing providers with technological leadership and scale advantages are poised to benefit from both customer diversification and rising per-chip value in this cycle.
Comments